Absolute Poverty: When Necessity Displaces Desire1

by

Robert C. Allen

Global Distinguished Professor of Economic HistoryFaculty of Social Science

New York University Abu DhabiSaadiyat Island

Abu Dhabi, United Arab Emirates

Senior Research FellowNuffield College

New Road Oxford OX1 1NFUnited Kingdom

email: bob.allen@nuffield.ox.ac.uk

June, 2017

1The main data sources for this article are the core price list for the InternationalComparisons Project 2011 and the regional price lists for Africa and Asia. I am grateful tothe World Bank and Dr. Nada Hamadeh for access and support. These data are available toresearchers upon application to the Director of the Data Development Group at the WorldBank. See ICP Global Office, 2011 ICP Data Access and Archiving Policy: GuidingPrinciples and Procedures for Data Access, 2012,http://pubdocs.worldbank.org/en/790491487172245068/121120-ICPDataAccessPrinciples-Procedures.pdf, pp. 6-7, for details on the application procedure. I am grateful to LigitaVisockyte for exceptional research assistance and to Mattia Fochesato for programmingassistance. I thank Anthony Atkinson, François Bourguignon, Shaohua Chen, Angus Deaton,Erwin Diewert, Stefan Durcan, Ian Gazeley, Rob Feenstra, Francisco Ferreira, PabloHernandez, Alan Heston, Sara Horrell, Dean Jolliffe, Branko Milanovich. Michail Moatsosas well as three referees and participants in several seminars for questions, comments,suggestions, and discussion.

2

Abstract

Robert C. Allen‘Absolute Poverty: When Necessity Displaces Desire’

A new basis for an international poverty measurement is proposed based on linearprogramming for specifying the least cost diet and explicit budgeting for non-food spending. This approach is superior to the World Bank’s ‘$-a-day’ line because it is (1) clearly relatedto survival and well being, (2) comparable across time and space since the same nutritionalrequirements are used everywhere while non-food spending is tailored to climate, (3) adjustsconsumption patterns to local prices, (4) presents no index number problems since solutionsare always in local prices, and (5) requires only readily available information. The newapproach implies much more poverty than the World Bank’s, especially in Asia.

JEL codes: I12, I32, O61, O63

keywords: absolute poverty, diet problem, linear programming, World Bank poverty line

Funding: This research was supported through a research allowance provided by New YorkUniversity Abu Dhabi.

The World Bank’s famous $-a-day poverty line began life as the finding of ascientific inquiry in the 1980s, became the Bank’s metric for measuring poverty in 1990, andreached full maturity when it was enshrined in the Millenium Development Goals as thestandard for tracking poverty around the world. However, the line rests on contestablefoundations that give rise to a host of theoretical and practical problems as well as leading,we argue, to underestimates of poverty in much of the developing world.2 While the WorldBank poverty line (WBPL) was originally conceived for developing economies, it must nowbe applied to all countries in view of the United Nation’s new Sustainable DevelopmentGoals, which came into effect on 1 January 2016. “Goal 1.1: By 2030, eradicate extremepoverty for all people everywhere, currently measured as people living on less than $1.25 aday.” (Atkinson 2017, p. 13). If extreme poverty is to be measured everywhere, then theWBPL must be revised since it is not valid outside of the tropics. This paper proposes a newmethod for defining absolute poverty that avoids the many problems of the World Bank’sline and which makes a more robust tool for measuring extreme poverty on a global basis.

The statistical origins of the World Bank Poverty Line (WBPL) give rise to many ofits difficulties. Ravallion, Datt, van de Walle (1991) were looking for a measure of absolutepoverty. They collected poverty lines for 33 countries ranging from the very poor to the rich,converted the lines to US dollars with a PPP exchange rate and plotted the lines against percapita consumption. They noticed that the lines of the six poorest countries were clusteredaround $1 per day in 1985 dollars, and that became their measure of absolute poverty. Thedollar value was raised to $1.08 in 1993 dollars when Chen and Ravallion (2001) updated thesame data with national price indices.

The empirical base of the WBPL was greatly strengthened in the new millenium whenRavallion, Chen, and Sangraulla (2009) put together a new data set of 74 national povertylines. Once again, poverty lines increased with income, but, in this case, no such trend wasapparent in the fifteen poorest countries.3 Ravallion, Chen, and Sangraulla concluded thatthese lines represented ‘absolute poverty’. Converting them to US dollars with PPPexchange rates produced the $1.25 poverty line in 2005 dollars. Most recently, the povertylines of these 15 countries in the own currencies were raised with national price indices to2011 values and then converted to US dollars with PPP exchange rates computed from the2011 round of the International Comparison Program (ICP 2011) price data. The averagecame to US$1.88, which was rounded up to $1.90, to give the current World Bank povertyline (Ferreira et al. 2016). Similar values have been reached by other methods (Kakwani andSon 2016, Sillers 2015, Joliffe and Prydz 2016). This ‘strange alignment of the stars’ hasincreased the line’s credibility (Atkinson 2017, pp. 19-20).

The line is only as good as these procedures, and they raise many troubling issues. First, which countries should be used to define ‘absolute poverty’? Should the referencegroup be updated? Deaton (2010, pp. 15-6) pointed out that India was in the original 1991sample of poor countries but grew so much in the next fifteen years that it was not in the2005 group of 15 poor countries. However, the Indian poverty line was very low, soexcluding it in 2005 raised the average and the WBPL with it. This meant that the number ofpoor in India and elsewhere increased markedly despite India’s economic growth–a perverse

2The line as been much debated. Recent contributions include Ferreira et al. (2016),Deaton (2010), Reddy and Pogge (2010), and Ravallion (2010). Related researchinvestigates subsistence lines (Lindgren 2015) and consumption floors (Ravallion 2016).

3The countries are: Malawi, Mali, Ethiopia, Sierra Leone, Niger, Uganda, Gambia,Rwanda, Guinea-Bassau, Tanzania, Tajikistan, Mozambique, Chad, Nepal, Ghana.

2

result, indeed!Second, measuring the prices is tricky. In some countries the data are nationally

representative; in others, they describe major cities only. In big countries like India andChina where regions are not well integrated, the relative prices for the whole country in theICP may not represent the price structure of any of its regions. The prices come fromsurveying shops, but in the case of small farmers, who eat some of their crop and sell the rest,the price of their food is the price they could have sold it for–not a price in a shop. And howshould we define the commodity? The ICP takes a very fine grained approach and tries tocompare rice to rice (indeed, the same grade of rice) and wheat to wheat. But if the poor eatonly wheat in a wheat growing country and rice in a rice growing country, might it not bebetter to compare the price of ‘grain’ and measure that as the price of wheat in the firstinstance and rice in the second? (Deaton 2010, p. 24) As it is, the ICP contains the prices ofmany commodities that have virtually no sales. The ICP 2011, for instance, reports the priceof mackerel in Zimbabwe, and that is an input into the food component of Zimbabwe’s PPPexchange rate, but mackerel is not eaten by the poor in this landlocked country and isprobably only available in a specialized shop in Harare.

Housing presents special problems. ICP 2011 contains rental (and purchase) valuesfor housing in some developing countries, and these are usually broken down alongtraditional/modern lines and whether or not they possess running water, electricity, indoorkitchens, and so forth. Location, which is of cardinal importance, is not explicitly dealt with,but presumably the rents are nationally representative. The problems in measuring thevolume and price of housing services across countries are so difficult that in 2005 the volumeof housing services in Africa and Asia was estimated as a constant percentage mark-up on therest of consumer expenditures. The implied prices of housing services were highly erratic(Deaton and Heston 2010, pp. 25-6). This approach is not systematic enough for povertymeasurement–especially when applied to rich countries.

Third, the usual index number issues surrounding formulas and weights bedevil boththe PPP exchange rates and measurement of inflation in national statistics. Conversions fromlocal currencies to US dollars are typically done with the PPP exchange rate for householdconsumption. The spending pattern of the poor differs from that of the average household:does that distort the result? Not as much as one would think (Deaton and Dupriez 2009). However, the spending pattern of the average household in Niger, for example, does differdramatically from that of the average household in the USA. If Törnqvist-Divisia indices areused, then the US shares are averaged with Niger’s to form the weights. These weights willbring into the calculation many goods and services that are never consumed by Niger’s poor(Deaton 2010, pp. 22-4).

Fourth, the existential meaning of poverty depends on the national poverty lines in thegroup defined as poor. What is their content? The poverty lines in the original RavallionRavallion, Datt, van de Walle (1991) study did not exhibit a common standard (Allen 2013).The sample of poverty lines for the 74 countries underlying the 2005 poverty line is puttogether more systematically. The usual procedure involves four steps. First, a calorierequirement is set. 2100 calories per person is often used, but, in fact, there is considerablevariation. Second, using data from a household expenditure survey for the country, find aband of the income or consumption distribution where average calorie consumption equalsthe chosen standard. Third, set the poverty line equal to the income or consumption per headfor the band choosing the calorie standard. The ‘food’ budget is what is spent on food andthe rest of spending is ‘nonfood’. No effort is made to investigate these aggregates further. Fourth, convert this national poverty line, which is in local currency, into US dollars with a

3

PPP exchange rate. Many features of this procedure are problematic. First, should a uniform calorie

standard be set or should it vary over time and across space (reflecting differences in workintensity) or demographic structure? In the event, a uniform calorie standard is notmaintained. Atkinson (2017, p. 128) found that it varied between 2030 and 3000 calories peradult in 10 poor countries, raising obvious questions of comparability. Second, how adequateis the rest of the food budget in terms of protein, fat, vitamins and minerals? Is it austere orluxurious? Third, how Spartan should the non-food budget be? How should it vary withclimate? Since most of the fifteen poor countries are in the tropics, it is unlikely that theaverage non-food budget includes enough fuel and clothing to survive a Russian winter, forinstance. Finally, even though the procedure is clear, it suffers greatly from a lack oftransparency. Since the Bank does not explore and assess what makes up ‘food’ and ‘non-food’ spending, there is no persuasive answer to the question ‘how can you live on $1 a day?’ Indeed, we show that people in the USA, the UK, and France could not live on the 2011version of that line.

The World Bank is committed to ending “chronic extreme poverty by 2030.” Toknow if it is succeeding, the Bank must measure poverty, and, in view of the manydifficulties in so doing, it “convened a high-level Commission led by Sir AnthonyAtkinson...to advise the World Bank on the methodology currently used for tracking povertyin terms of people’s consumption, given that prices change over time and purchasing powerparities across nations shift.”4 While Atkinson was aware of the many difficulties with of thecurrent poverty line and favoured further research to improve it (including the approach ofthis paper), he never-the-less endorsed the continued use of the local currency equivalents ofthe $1.90 line with the only adjustment being to raise them over time in line with inflation inlocal prices. There are two reasons for this approach. First, Atkinson’s terms of referencewere narrow. “The Commission was asked to take the 2015 estimates as its point ofdeparture and to assess how the process may be carried forward to monitor progress up to2030 in achieving SDG goal 1.1.” Second, “the $1.90 [line] has acquired an independentpolitical status.” Better to stick with imperfect goal posts rather than muddle the scoring byshifting the goal posts in the middle of the match. (Atkinson 2017, pp. 22, 30, Ferreira et al.2016.)

While Atkinson’s recommendation makes administrative sense, it does not resolve theunderlying scientific and philosophical difficulties. Ideally, the international poverty lineshould satisfy five criteria: (1) It should have a clear meaning related to survival, health, andwell being. In terms of Sen’s (1987, 1992) theory of capabilities, the poverty line shouldsustain basic functionings like being active, growing, and healthy. The line should also (2)represent a constant standard across time and space, (3) respond to local prices and otherpertinent local factors like climate, (4) avoid intractable index number problems, and (5)require only readily available information.

This paper develops an approach that satisfies these requirements. The paper operateson both the normative and the positive plane. On the normative plane, I propose that thepoverty line be set by explicitly budgeting for basic needs. This is a long standing traditionthat is still widely used in Europe and the USA when ‘reference budgets’ are drawn up tospecify poverty lines (Rowntree 1901, Goedemé et al. 2015, Carlson et al 2007.) The Basic

4These quotations are from the Forward to the report by Kaushik Basu, ChiefEconomist and Senior Vice President of the World Bank Group (Atkinson 2017, p. vii).

4

Needs Poverty Line (BNPL) developed here has three categories of spending–food, non-foodgoods, and rented housing. My approach uses linear programming to set the diet portion ofthe poverty budget and early twentieth budget studies for industrial workers in St. Petersburgand Bombay to make non-food spending climate dependent. Housing is also explicitlybudgeted. This approach avoids all of the problems of the World Bank’s methodology. Iillustrate the method here using a sample of twenty countries ranging from Niger andZimbabwe to France, the UK, and the USA.

An objection to linear programming diets is that they are unrepresentative of actualbehaviour and so of no use in setting a poverty line or for any other practical purpose. Thesecond plane on which the paper operates is positive and aims to assess the predictive powerof linear programming diets. I agree with the conventional view (Stigler 1945) that linearprogramming utterly fails to explain the diets in rich countries. It is a different story indeveloping countries, however. Linear programming does a reasonable job in explaining thetotal quantity of food and its distribution among broad categories. The linear programmingdiets are predominantly vegetarian. Grain bulks very large. Small quantities of animalproteins and oil are also consumed. Linear programming usually predicts the predominantgrain consumed in a country. It cannot predict the variety of fruits and vegetables that thepoor eat, nor can it explain the universal consumption of small quantities of sugar nor eatingassociated with festivals. There is a small amount of ‘wiggle room’ in which people havesome space to consume a little sugar, favourite spices, or traditional foods at the expense of ahealthy diet. But it is the nature of poverty that the latitude for these substitutions islimited–certainly in comparison to rich countries.5

Specifying the poverty line diet with linear programmingThe diet problem was the first linear programming problem ever formulated in a

famous paper by Stigler (1945). The problem is to choose a diet from a list of foods thatminimizes the cost of meeting a set of nutritional requirements. The objective function to beminimized is the cost of the diet:

Cost = 3piFi (1)

where pi is the price of a food and Fi is the quantity of the food consumed. The summationcan extended over a list of t foods F1...Ft, many of which will not selected in the solution.

The nutritional requirements are specified with a set of inequalities, each of whichsets the requirement for one nutrient.

3njiFi $ Rj (2)

Here the summation also runs over all of the foods indexed with i. Rj is the required amountof nutrient j–the minimum calorie requirement, for instance. nji is the quantity of nutrient jper unit of food Fi, for instance, nji might be the number of calories per kilogram of wheatflour and Fi, the kilograms of wheat flour in the diet. Each nutrient required in the diet has aninequality describing that requirement.

Finally, the consumption of each food has to be at least zero:

5This is a specific sense in which economic development creates freedom (Sen(1999).

5

Fi $ 0 for all i (3)

The linear program model presents a neat contrast to the standard model of consumerchoice in its dual form. In that form, consumer goods are chosen to minimize the cost ofmeeting a specified utility level U*. The difference with the linear programming model isthat the inequalities (2) are replaced by the utility constraint, equation (4) :

U(F1...Ft) $ U* (4)

This is the formal sense in which necessity displaces desire in the definition of absolutepoverty.

These days, linear programs can be easily solved with the simplex algorithm in Excel. The solutions have two properties that are important for defining the poverty line. First,increasing the number of requirements or increasing the magnitude of a requirement eitherleaves the cost of the diet unchanged or increases it. A more nutritious diet is never cheaperthan a less nutritious diet and may well cost more. Second, the maximum number of foodsthat solves the problem is equal to or less than the number of requirements. The number ofrequirements, therefore, limits the variety of the diet.

In his original investigation of the diet problem, Stigler (1945) used US prices from1939 and 1945 to compute the cost of the least cost diet meeting a set of requirementsincluding calories, protein, iron, niacin, calcium, vitamin C, vitamin A, thiamine, andriboflavin with values appropriate to a ‘moderately active’ man weighing 154 lbs. Stigler didnot have Excel at his command but nevertheless reasoned his way to almost the correctanswer. The solution for 1939 was 168 kg of wheat flour, 129 kg of dried navy beans, 23 kgof evaporated milk, 50 kg of cabbage and 10 kg of spinach. The values warrant comparisonwith ones we compute for developing countries in 2011.

Stigler’s reaction to the solution has also been important: he thought the diet wasimpractical. “No one recommends these diets for anyone, let alone everyone; it would be theheight of absurdity to practice extreme economy at the dinner table in order to have an excessof housing or recreation or leisure.” (Stigler 1945 , pp. 312-3) This theme has been taken upby subsequent economists, who have tried to incorporate ‘palatability’ into the program. Smith (1959, p. 272) remarked that Stigler’s diet was “a dramatic illustration of how littlepurely nutritional needs have to do with the level of actual food expenditures...If we wantdiets that someone might be willing to eat, we need models that take account of tastes andhabits.” This is surely true of people in rich countries whose behaviour is determined bypreferences, income, and prices. Linear programming is much more germane to poor people,however. For them, survival is the issue, and the needs for survival take precedence. Preferences and income give way to nutritional requirements in determining consumptionwith prices still playing a role. Indeed, from the linear programming perspective, what itmeans to be poor is that your life is governed by linear programming, rather than standardconsumer theory.

‘Nutritional requirements’ has an aura of scientific objectivity, and Stigler (1945) andSmith (1959) adopted lists of requirements issued by nutritional boards without criticism orexamination. Indeed, it was the desire for an objective standard for poverty that motivatedthe research described here. However, it is clear on examination that the nutritionalrequirements set by bodies like the World Health Organization are in important respects

6

subjective.6 First, precise values for some nutrients such as calories and protein can bespecified with reasonable accuracy, but for others that is not possible. Niacin, for instance, isnecessary to prevent pellagra, and field observations suggest widespread appearance ofpellagra in populations where adult men receive less than 7 mg of niacin per day. However,the current WHO requirement for adult men is set at 21 mg on the grounds that the highervalue contributes to better health (Prinzo 2000, p. 11, Rao 2009, pp. 256, 260). The povertyline distinguishes the ‘poor’ from the ‘non-poor’. Should the line be set at 7 mg or 21 mg orsomewhere else? I have adopted the recommendations of the medical authorities, but theuncertainties should be recognized. Second, for this reason, most of the world’s population isdeficient in some nutrients. 90% of the Indian population, for instance, is anaemic by currentstandards, which means they are deficient in iron, thiamine, or folic acid. Evidently, many‘non-poor’ are deficient in these regards, so that full adequacy with respect to iron does notdistinguish the poor from the non-poor. Perhaps ‘moderate anaemia’ should be the dividingline with correspondingly reduced nutritional requirements? Third, for geographical reasons,some required nutrients are not available to most of the world’s population. Iodine, forinstance, is naturally available only to people living near oyster beds. Unless iodized salt isavailable, most people in the world would be iodine deficient, so there is no point including itas a requirement in a programming model defining a poverty line. Fourth, none of thesestandards takes into account the seriousness of the impairment to life that results from thedeficiency. It may be that most people are unconcerned about vitamin A deficiency becausenight blindness does not appear a costly disability–at least not sufficiently detrimental torequire the expenditure necessary to eliminate it.

These uncertainties affect the linear programming approach to diet in two importantways. First, we omit from consideration nutrients whose availability are locationallyspecific. Iodine is an example, as is vitamin D. People are not vitamin D deficient in sunnyclimates, although they may be deficient in cloudy, wet places. The poverty line is meant todistinguish the poor from the better off, and the availability of iodine and vitamin D does notdo that.

Second, with respect to other nutrients, the linear programming approach takes on thecharacter of an estimation exercise rather than a purely objective determination of the optimaldiet. One question we ask is whether there is a set of nutritional standards that is commonacross the world and that rationalizes the diets that poor people consume. We argue that theanswer is (approximately) yes, and that is the standard incorporated in our measure ofpoverty. In this sense, the choices made by the poor imply the poverty line: The poor have avoice in defining poverty–even if they are not aware of it.

Data and empirical specification

To compare our results to the World Bank’s poverty line of $1.90 per day in 2011, weneed prices from 2011. The principal data source is the ICP2011 core spreadsheet and theregional spreadsheets for Africa and Asia.7 The ICP is a tremendous achievement, but it was

6For a list of relevant WHO publications, most available online, seehttp://www.who.int/nutrition/publications/nutrientrequirements/en/

7The core prices were taken from ICP2011: Data for Researchers, the African pricesfrom ICP2011_AFR_Regional2011, and the Asian from ICP2011_ASI_Regional2011. I am

7

necessary to fill some lacunae and add accommodation rental prices derived from othersources (See Appendix). I investigate the implications of these prices for 20 countriesranging from the poorest to the richest (Table 1 onward).

Countries are a natural unit of analysis since poverty is a national political issue andsince countries share a common currency, but there are also reasons for choosing a regionalapproach. Prices differ across regions in large countries with poor infrastructure or otherimpediments to trade, and those differences imply different poverty lines. Climate also varieswithin countries. The World Bank has set different lines for urban and rural parts of India,China, and Indonesia. However, it is not possible to explore these issues using the ICP,which is the basis for international comparisons, since it reports only one price for each goodin each country. When regional issues look likely to be important, they will be noted.

least cost diets: 1700 calorie model

We begin by examining the diets implied by various nutritional requirements.8 Weconsider them in an increasingly stringent progression. There are four models in thesequence. Each contains all of the nutrients of the previous step and increases the quantity ofthose nutrients or adds additional nutrients or both.9 The models are:

! 1700 calorie model. The only requirement is 1700 calories per day.! CPF model. Three nutrients are required: 2100 Calories per day, 50 g. of Protein,

and 34 g. of Fat! basic model. CPF requirements plus the Indian recommended daily allowances

(RDA) of iron, folate, thiamine, niacin, and vitamins C and B12.! full course model. Basic model plus RDA of six more vitamins and minerals.

We begin with the most elementary requirement–calories. What is the minimum costof a diet that supplies just enough calories for survival? By ‘survival’ we do not mean theminimum for a single adult to subsist from one day to the next but rather the minimum, onaverage, for the species to survive. Adults must have enough energy to work and children togrow.

The minimum society-wide average can be established in two ways. One is bycalculation.10 The distribution of the population by age and sex is determined, and the energy

grateful to Nada Hamadeh and the World Bank for making these data available to me.

8Each requirement is expressed as an inequality in the form of equation (2). Thequantity of each nutrient per kilogram of food (nji) must be specified. Generally the valuesused were those shown on the US Department of Agriculture, National Nutrition Databasewebsite. Some values, however, were taken from the regional nutritional databases listed inthe online references. These data bases often do not agree with each other, and it might beimportant to investigate these discrepancies, but that has not been done here.

9Details about the linear programs are found in the appendix ‘notes on the linearprogramming.’

10FAO (2001) explains the methodology.

8

required for basal metabolism for each age-sex group is calculated with standard formulae. The results depend on the average height of each group and the Body Mass Index that each isexpected to maintain. Additional allowances are also required for pregnant and lactatingwomen. Basal metabolism of each group is then increased by its physical activity level(PAL). Determining the PAL requires constructing an activity schedule across the year, sothat the appropriate mark-up can be applied to each hour (the physical activity ratio or PAR)depending on exertion. More strenuous activities get higher PAR’s. The PAL can then becomputed as the average of the PAR’s over the year.

Calculations along these lines point to around 2000 calories per person per year as theaverage requirement. This provides enough for some people to work very hard and forchildren to have enough energy to grow. The requirement varies depending on the agedistribution of the population: faster growing populations have more children and a loweraverage calorie requirement. Calculations by the Food and Agricultural Organization (2008,p. 8) indicate a requirement of 1600-2000 calories per person per day. The US Departmentof Agriculture (2010, p. 2) uses a standard of 2100 calorie per person per day (with anunspecified variation across regions) in assessing food security. I have computed the samefigure for Britain in 1841 assuming that the average man was a carpenter and the averagewoman a domestic spinner (Allen 2013).

An issue that arises with respect to energy requirements is the question of whether theshift from farm work to urban work as well as the mechanization of farm tasks has led toreductions in the need for calories (Deaton and Drèze 2009, pp. 57-8). Probably not bymuch. The model of Britain in 1841 can be used to calculate bounds by comparing theaverage energy requirement (a) if all men performed strenuous work on a continuous basiswith the requirement (b) if half performed moderate work and the other half light work. Energy consumption averaged over the population drops by about 250 calories per day. Thisdrop must overstate the actual change since not all men did strenuous activity all of the timein the ‘olden days,’ but it is difficult to explore this further in the absence of detailedinformation on work intensity and how it has changed. I have made no adjustment for thiseffect.

The second approach to determining calorie requirements is to look at what peopleactually consume. Survey data for India shows that the poorest decile of the populationconsumes about 1450 calories per person per year (Deaton and Drèze 2009, p. 47,Suryanarayana 2009, p. 35). This is below basal metabolism, so it is either an error, or itindicates an unusual demographic structure (which means it is not relevant for society as awhole), or the population is dying out (in which case the standard is too low).

The second decile from the bottom consumes on average 1700 calories per person perday (Suryanarayana 2009, p. 35). This is just above the lowest FAO value and about thebare minimum a group requires for survival.

In view of these considerations, linear programming diets were calculated with theonly constraint being 1700 calories per person per day. The implied diets are in Table 1.With only one constraint, there can be only one food in the solution to the programmingproblem. For 12 countries that is a cereal or flour (170-178 kg/year depending on the kind orabout a pound per day). For the other eight, it is vegetable oil (70 kg/year or about one cupper day). These are small quantities, which suggests that people eating them might behungry. The solutions of the linear programs are in kilograms of the various foods, and thetotal provides a rough indicator of nutritional intake that we use to gauge the effect ofincreasingly stringent nutritional requirements. Total weight of a diet also provides asummary statistic to compare predicted consumption with actual diets.

9

The appearance of oil is unexpected, and it is probably also a recent phenomenon inworld history. It reflects the widespread cultivation of palm oil in south Asia, a developmentof the late nineteenth century. Before that, rice or some other grain was the cheapest sourceof calories around the Pacific Ocean.11

It is a tricky question whether man can live by maize alone, but surely he cannot livesolely on vegetable oil. Aside from fat, it supplies no nutrients. A population could notsurvive on the vegetable oil diet. Requiring only calories leads to death rather than survival.

Least cost diets: CPF model

More satisfactory diets are implied by imposing more requirements. The second classof requirements are the principal nutrients–calories, protein, and fat. In the calculations, weincrease the calorie requirement to the USDA value of 2100 per day. This allows people amore ample supply of energy to do the work that sustains society as well as raising children. Protein is set at 50 grams per person per day. The ultimate basis of this value is experimentsthat measure the nitrogen intake required to match the body’s excretion of nitrogen and thusto maintain the body’s nitrogen stocks. Fat is set at 34 grams per person per day, the amountthat supplies 15% of the energy intake (FAO 2008c, p. 14). These requirements depend onbody mass, age, sex, pregnancy, lactation, and so forth. In these cases (as with all othernutrients to be considered), the value of the requirement used in the linear program iscalculated from age and sex specific requirements as a society-wide weighted average usingthe age and sex distribution of the Indian population as weights.12 Recommended values forIndia are used, as they are more likely to reflect conditions in developing countries todaythan global recommendations.13 The protein requirement lies at about the 30th percentile ofthe Indian income distribution, while the fat requirement is in the middle (Suryanarayana2009, pp. 35-6).

Tables 2 and 3 summarize key features of the diets as functions of the nutrientrequirements. Table 2 shows average annual food consumption in kilograms. With the 1700calorie diet, the average was 131. This increased to 200 kilograms with the CPF diet. Thenumber of foods in the diet also increased (Table 3). There was only one food chosen with

11Linear programs like those reported in this paper have been run with price datacollected by Lockyer (1711, pp. 148-151) in Canton in December, 1704. Rice rather than oilwas the solution to the 1700 calorie model, indicating that oil was a more expensive source ofcalories than rice.

12The nutritional requirements are from Rao (2009, pp. 105-6, 124, 119-32) and thepopulation structure from http://esa.un.org/unpd/wpp/Excel‐Data/population.htm There ismore variation in the recommendations of professional bodies regarding fat than for othernutrients, and I have opted for a low requirement.

13The RDAs include an allowance for losses during cooking. “Considering thecooking loss of 50%, the RDA of ascorbic acid has been set at 60 mg/day.” (Rao 2009, p.287). An advantage of using Indian RDAs is that the cooking losses are assessed in terms ofIndian culinary practices, which are probably more representative of developing, tropicalcountries than the cooking practices in the West. See Rao (2009, pp. 14, 246, 248, 251, 257,262, 272, 274, 275, 279, 286) for more examples.

10

the 1700 calorie diet. The linear programming solution allows up to three foods with theCPF diet. Three foods are chosen in eight cases and two foods in twelve for an average of2.40 foods.

The diets that solve the linear program with the principal nutrients as constraints areshown in Table 4. Consumption of oil is cut dramatically to plausible levels, and some isused everywhere. Wheat is the staple in wheat growing areas, as is rice in southeast Asia,and millet, sorghum, or maize in sub-Saharan Africa. Legumes are consumed in six of thecases including all of the rice based diets. It is significant that the diets are purely vegetarian,and that no alcohol, sugar, or vegetables (other than the legumes) are consumed. There is nosugar, no alcohol, and very little meat in any of the diets implied by linear programming.

Least cost diets: basic model

While the CPF diets provide better nutrition than the 1700 calorie diet, they none-the-less suffer many deficiencies. We begin with those that could lead to four of the mostcommon and serious deficiency diseases. Pellagra is due to insufficient niacin, beri-beri to alack of vitamin B1, scurvy to insufficient vitamin C, while anaemia can be due to inadequatelevels of either iron, thiamine, or folate (folic acid). Table 5 reports nutritional consumptionrelative to recommended daily allowances for these nutrients in the CPF diet.

In most cases, the CPF least cost diets meet the requirements for calories, protein, andfat exactly, and, when the requirements are over fulfilled, the excess is minor. So far as theminerals and vitamins are concerned, the diets supply no vitamin B12–this is found only inanimal products–and none or only negligible quantities of vitamin C. The absence of vitaminB12 means that anaemia would be widespread unless consumption of B12 were inadvertent. In India, “since populations subsisting essentially on foods of vegetable origin do not showevidence of widespread vitamin B12 deficiency, it is speculated that polluted environmentand unhygienic practices could be providing the necessary minimal vitamin B12.” (Rao2009, p. 278) The lack of vitamin C implies widespread scurvy.

There is a likelihood of other deficiency diseases as well. Two kinds of diets areparticularly deficient. The first are the rice-based diets deduced for Vietnam and Myanmar. These diets have low enough niacin levels to suggest wide-spread pellagra and low B1 levelsindicating a risk of beri-beri. It is significant that the short-grain, milled rice which theyconsume is particularly lacking in these nutrients. In contrast, the brown rice consumed inSri Lanka supplies more niacin and thiamine, so the deficiency problems are not so severe.

The second kind of diet that indicates a likelihood of deficiency diseases is the wheat-based diet of France, Algeria, and Lithuania. Refined wheat flour in these countries is notenriched, so it lacks niacin and thiamine. Otherwise similar diets in the USA, UK, Turkey,and Mexico do not lead to these inadequacies because the enrichment of wheat flour ismandatory. The comparison indicates the benefits of mandatory food fortification.

In terms of the linear programming, the deficiencies can be cured by imposing therequirements on the solution. As noted previously, we compute the requirements from theIndian recommended daily allowances by computing the weighted average of the RDAs forthe various age and sex groups. The results are shown in Table 6.

The additional requirements have major implications for the linear programmingsolutions. The first is that the volume of food consumed over the year goes up from 200 kgwith the CPF diet to 344 kg. More food gives more nutrients.

The second change is an increase in the number of foods from 2.40 on average in theCPF diet to 5.00. The increase is due mainly to the addition of an animal product and a

11

vegetable. Animal products enter the solution as a consequence of requiring vitamin B12. The linear programming solution generally implies either the cheapest available fish–usuallymackerel–in coastal districts or milk in inland regions. Meat in any form rarely appears inthe solution to a linear program. The appearance of vegetables–most commonly cabbage–orcassava is due to the vitamin C requirment. The B12 and C requirements are independent ofthe others, so adding these requirements to the program has scant impact on the rest of thediet. Qualitatively, the pattern of food consumption is similar in the CPF and reduced basicmodel. The same grains are generally consumed in the same regions. Total foodconsumption rises because of the introduction of animal protein and vegetables.

The increase in total food consumption has another implication that becomesincreasingly important, namely, the overshooting of requirements. With the CFP diet, mostsolutions meet the calorie, protein, and fat constraints exactly. The average degree ofovershooting is only a few percentage points. Overshooting is more widespread with thebasic diet. Three of the solutions overshoot calories, and the average excess is 7%. Virtuallyall of the diets overshoot protein, and the average excess is 32%. Most solutions meet the fatrequirment exactly, but the requirement is nonetheless exceeded by an average of 13%.

full course model

The vitamins and minerals considered thus far are only a subset of all of the nutrientsthat might be considered. Recommended daily allowances have been set for many others. To explore the implications of some of these, requirements for vitamin A, B6, riboflavin,calcium, magnesium, and zinc have been added to the linear program.

The exercise has a surreal air because of difficulties in defining and assessingdeficiencies. In the case of vitamin B6, for instance, it is difficult to measure the extent ofdeficiency in the population (Rao 2009, pp. 260-3). Setting RDAs is difficult in some cases(vitamin B6) and fraught with conflicting considerations in others. Calcium requirementsdepend on vitamin D and protein intake. Low protein consumption reduces calciumrequirements meaning that standards set for rich people are too high for poor people, and bysome measures most Indians look like they get enough calcium. On the other hand, femurfractures occur at younger ages amongst poor women in India suggesting there might be anissue about calcium adequacy after all. (Rao 2009, pp. 158-9) At what level should thecalcium RDA be set? In other cases, it is not clear how serious the deficiencies might be. Alack of vitamin A leads to night blindness, but how costly is that? (Rao 2009, p. 296) Inother cases, deficiencies are so common or so rare that the intake of the nutrient provideslittle information about poverty or wealth. Thus, “dietary deficiency of riboflavin is rampantin India...only about 13% households meet the dietary requirement.” (Rao 2009, p. 251) Incontrast, “the available reports...in India...do not report any widespread zinc inadequacy.” (Rao 2009, p. 225). In neither case, does the RDA provide a boundary that distinguishespoor people from better off people.

Introducing these additional vitamin and mineral requirements implies increased foodintake. The number of foods in the diets rises from an average of 5.00 with the basic diet to6.4. In addition, more nutrients are obtained by increasing the quantity of food consumed ina year from an average of 344 kg with the basic diet to 426. The increase is greatest amongthe developing countries where average food consumption rises to 440 kg. The Chinese dietreaches a staggering 763 kg. Increasing the volume of food to this extent leads toconsiderable overshooting of calorie requirements (by 10%) and especially proteinrequirements (by 39% on average).

12

Comparison of the details of the full course diet (Table 7) to the basic diet (Table 6)shows some unusual changes. The consumption of wheat flour falls, while the consumptionof legumes, cassave, vegetables and potatoes reach extreme limits. These features raisequestions about the empirical relevance of the diets.

The full course diets have affinities with Stigler’s original linear programming diets. The same foods turn up, and the quantities are of similar magnitudes. The reason is thatStigler’s nutritional requirements include calories and protein as well as most of the vitaminsand minerals considered here. Stigler’s specification did not include a fat requirement, andits absence explains why there is no oil or butter in his solution.

Linear programming diets and the diets of the poorStigler warned us that linear programming diets provided no guidance for the

behaviour of Americans, and indeed, the solutions he found do not described what Americansate in the 1930s and 1940s. Does this judgement apply to people in ‘absolute poverty’ indeveloping countries today? With some qualification, the answer is no. Linear programmingexplains many features of their behaviour.

We have examined a range of four linear programming solutions reflecting differentlevels of required nutrition. Not all of these explain behaviour. For many countries, a purevegetable oil diet was the solution for the 1700 calorie diet. That diet cannot sustain life andno one consumes it. The CPF diet looks more promising, but it does not include any animalproducts or any fruits, cassava, or vegetables (with the exception of beans in the rice baseddiets). Most vegetarians consume dairy products or eggs, and fruit and veg are almostuniversally consumed, as will be shown, so the CPF diet does not describe human behaviourwell. At the other extreme, the full course diet predicts too much food consumption. 440 kgper year greatly exceeds per capita consumption in much of the world in the 1960s and sodoes not describe what the poor were eating. We are left with the basic model as the bestcandidate for describing behaviour.

We can compare the predictions of the basic model with food consumption to assessits merits. Ideally, one would test the model against the spending patterns for the incomeband defining the poverty line in each country in 2011. This information is only available fora few countries and never for 2011. The most widely available information is averageconsumption as summarized by the UN FAO Food Balances Sheets. I use the balances for1961, the earliest year available, when many people in developing countries wereundoubtedly poor. The first paper to measure poverty globally was Ahluwalia, Carter, andChenery (1979, p. 4), which set the poverty line at consumption of the 45th percentile in Indiain the 1970s. This was not far off average consumption in 1961.

Table 8 compares the predictions of the basic model with the 1961 consumptionpattern for nineteen countries (data are not available for Lithuania). The correspondencebetween prediction and behaviour is much higher for the developing countries than for thericher countries. Consider first the total weight of the diet. In the rich OECD countries,actual was 2.89 times predicted; in the middle income OECD countries, the correspondingratio equalled 1.90, while among the eleven developing countries, actual was only 14%greater than predicted.

LP models also work best for developing countries when the components of the dietare considered. In the developing countries, the model predicts that animal products wouldamount to 10% of consumption, while average consumption in 1961 was 12%. For fats, theprediction was 2% against a 1961 value of 1%. LP over predicts the consumption of grainand grains–58% against a reality of 41%–with the discrepancy made up with greater

13

consumption of vegetables and fruits–prediction of 31% versus actual consumption of 40%and ‘other product’ where the model predicts zero and 1961 consumption was 6%. The latterincludes sugar (average of 9 kg per head), which was consumed everywhere, and alcohol(average of 8 kg), which was consumed in significant quantity only in Zimbabwe andGambia (54 and 40 kg per head respectively).

These predictions are not perfect but are far closer to reality than the predictions forrich countries. For the USA, UK, and France, the prediction is that grain and grain productswould amount to 65% of consumption, whereas the actual value in 1961 was 12%. Thediscrepancies was matched by under predicting the consumption of animal products (15%versus a reality of 40%), fruit and vegetable consumption (16% versus 29%) and theconsumption of ‘other products (zero versus 18%). The latter consisted mainly of sugar (45kg per year) and alcoholic drinks (112 kg per year). The big errors for the rich countriesillustrate Stigler’s point about the failure of linear programming to predict behaviour, whilethe much smaller errors for developing countries illustrate my point that linear programmingcan predict fundamental features of the behaviour of the poor.

How well does linear programming perform within the broad categories? Linearprogramming performs poorly in explaining the variety of foods eaten. This is especiallytrue of fruits and vegetables. Normally, linear programming selects cassava or a singlevegetable, which is generally the cheapest source of vitamin C. In reality, people consumemany different fruits, nuts, and vegetables. Some of this behaviour is a response toseasonable prices changes, which do not appear in the ICP. A fundamental reason, however,is that many vegetables sell at similar prices and have similar quantities of vitamin C. Thediet can be diversified at very little cost. The same is true of some fish and animal products.

On the other hand, the most important food consumed by the poor is grains and grainproducts, and linear programming correctly predicts the most important grain in eleven of thefourteen developing countries in the sample. Success is high in Africa (millet and sorghumin Niger, maize in Zimbabwe, rice in Liberia, wheat in Algeria and Egypt with some successin predicting grains of secondary importance) and south Asia (Bangladesh, Myanmar,Indonesia, Sri Lanka, and Vietnam all rice).

The model is off the mark in three cases. I tally Thailand as an error since maize isthe grain in the linear programming solution, but rice was the predominant crop in 1961. Inthat year, very little maize was consumed by humans, but now it comprises 10% of grainconsumption, presumably eaten by the poor. Linear programming anticipated thisdevelopment, so perhaps this is not a prediction error at all.

The predictions are most problematic for the biggest countries, China and India,where the model predicts sorghum and millet as the principal grains. In India these cropscomprised 22% of grain consumption in 1961 (down to 9% in 2011) and weredisproportionately eaten by the poor. The thing is, however, that the poor also consumedlarge quantities of more expensive wheat and rice, and this behaviour is not captured by thebasic model. The situation was similar in China where millet and sorghum comprised 18%of the grain consumed by humans in 1961. Jinqing (2005, pp. 85, 124, 146-7) discusses thewidespread consumption of sorghum and other coarse grains by peasants in Henan in 1996,for instance. (Virtually none is eaten today.)

The results for India and China raise several broad questions. The first relates to theICP data used here. The ICP includes a single price for each item that is supposed to berepresentative of the whole country. India and China are very large, and their economies nothighly integrated. The relative prices in the regions differ from the relative prices shown inthe ICP. In that case, it would be better to apply the linear programming approach to sub-

14

national units, so the baskets would differ in wheat growing districts and rice growingdistricts, for instance. The same criticism applies to any other approaches (including theWorld Bank’s) that come up with a single basket for large and diverse countries.

The second issue relates to the importance of habit in diet choices. An extreme viewwould maintain that food choices are entirely driven by customs that have nothing to do withwhat is cheap. That view is hard to credit for developing countries in view of the results justreported. Atkin (2013, 2016) has proposed a more specific hypothesis that warrantsattention. He claims that preferences do reflect local scarcities and these preferences acquirea life of their own, so that they continue to influence behaviour when market integrationincreases or people migrate to districts where relative prices differ. Migrants continue topurchase the foods to which they are accustomed with the result that their diets cost morethan those of their neighbours whose tastes accord with the scarcities of their new homes. Atkin’s (2013, p. 1147) analysis of intra-Indian migration suggests that habit raises the costof calories by about 5% even for deprived groups. A 5% increase in the cost of food wouldincrease the poverty line by about 3% given food’s share in expenditure. This is not a largeamount in view of the many other uncertainties involved in setting the poverty line. Furthermore, habits decay over time, and people adapt to the new price environment.

Third, there is a question: should we pay attention to food habits in setting the povertyline? The former is a matter of fact, while the latter is a matter of value. Many Indians preferrice and wheat to millet and sorghum. People in Niger eat nothing but millet and sorghum. Should Indians be given a more costly standard than people in Niger just because they have ataste for more expensive food? The poverty line represents the cost of meeting basic needs,not a level of satisfaction, and should be set accordingly.

On the practical plane, however, it should be noted that the linear programmingpoverty line is generous enough to allow for a small accommodation of habit or variety indiet. This latitude arises because the requirements for micro nutrients are set at levels that themedical profession judges to be needed for healthy living. These levels exceed intake neededto prevent acute deficiency symptoms, as we have mentioned. People who receive theincome to consume the recommended daily allowance of some nutrient might purchase a dietwith less than the RDA in order to indulge a taste. This is probably the reason that most ofthe Indian population is anaemic, for instance. Likewise a calorie standard of 2100 is higherthan absolutely necessary for survival, so taste can be indulged at the expense of energy. Poverty is constraining and linear programming captures those bonds, but it is not whollyimmobilizing.

Non-food consumptionA poverty budget includes items besides food. In the World Bank approach, non-food

expenditure is the average mark-up of non-food spending in the poor countries whosebudgets underlie the line. There is no reason to suppose that these mark-ups reflect identicalnon-food budgets–and, indeed, the claim that “the judgements made in setting the variousparameters of a poverty line are likely to reflect prevailing notions of what poverty means ineach country setting” (Ravallion, Chen, Sangraula 2009, p. 167)–belies that possibility. Since the poor countries are mainly tropical, the only safe assumption is that the non-foodspending is appropriate to tropical conditions.

I address this limitation by setting an explicit non-food budget. It is intentionally

15

austere and is limited to housing, fuel, lighting, clothing, and soap.14 The cost of education,medical care, and so forth are not included, so the resulting poverty line is an absoluteminimum. Arbitrariness is unavoidable. Our approach makes the decisions visible, so theycan be debated, rather than leaving them unexamined under the rubric of “other spending.”The linear programming framework could be extended to include the non-foods byspecifying the requirements in terms of square metres of living space, BTUs of energy, etc.,but the linear programming problem decomposes into separate problems so long as the goodsin each category contributed only to meeting the requirement of that category, which is themaintained assumption. So I analyse the categories in turn.

I set the quantity of housing at three square metres per person. By the standards ofrich countries, this represents extreme–and often illegal–overcrowding. Even illegallysubdivided apartments in New York offer 5-10 sq metres per person (Gadanho 2014, p. 135). Exceptionally high densities, however, are common in Third World slums. In Bombay in1921, for instance, cotton mill workers lived one family to a room of 13.3 square metresgiving each person 2.3 sq metres of space. In Ahmedabad in the 1920s, the average was 3.6square metres per head, in Shanghai in 1952, the average resident had 3.4 square metres, andin the slums of Nairobi today, the rate is 3 sq metres per person. UN Habitat (2003)proffered a definition of overcrowding as more than 2 people per room or less than 5 sqmetres each. In 2010, however, this standard was revised to 3 people per room, implying lessthan 3.3 square metres per person.15

To work out the cost of 3 square metres, we need the rent per square metre. For sixpoor countries (Algeria, Niger, Bangladesh, Indonesia, Sri Lanka, and Vietnam), the ICPreports rents for ‘traditional’ structures, so rent per square metre can be calculated.16 For theremaining countries, it was necessary use other sources detailed in the data appendix. Theaim was to find market (not subsidized) rents in the poorest districts of the capital or otherlarge cities. In some cases, like New York, rents were found for illegal, substandardapartments, although, in this case, the rent per square did not differ greatly from the USDepartment of Housing Fair Market Rent for New York County in 2011. Table 10 tabulatesthe rents. The differences across countries are striking. In most of the poor countries, rentswere $.50 - $1.00 per square metre per month. Even in China, where the rent is the freemarket average for the eight largest cities, it came to only $1.28 per square metre. On theother hand, in the USA, UK, and France, rents were on the order of $25 per square metre permonth. In the USA, rent on 3 square metres worked out to be $1.76 per day, which shows theimpossibility of living on $1.90.

The requirements of clothing, fuel, and lighting depend on climate. A point ofdeparture for fuel and lighting is the Energy Poverty Line of the Millenium Development

14The soap requirement was arbitrarily set at 1.3 kilograms per year (25 grams perweek).

15Shirras (1923, p. 25), Bombay Labour Office (1928, p. 19), Sun, Linyun (2011, p.57), UN-Habitat (2003a, p. 237), UN-Habitat (2003b, p.12), UN-Habitat (2010, p. 16).

16The rents shown in this paper are averages of at least two housing classes. Theygenerally have electricity and indoor water supply. Rents for poorer dwellings lackingelectricity and indoor water or for better dwellings, which also have indoor toilets andkitchens, give very similar results in these international comparisons.

16

Goals, which sets the minimum at 1.6 million BTUs of fuel and .4 million BTUs for lighting(Modi, McDade, Lallement, and Saghir 2006, p. 9). The former, which is based onengineering studies, provides enough energy for cooking but nothing beyond that for heating,so the requirement is suitable only for hot climates. The latter provides enough energy forthree hours of lighting per night from a candle or an electric light bulb. Other sources ofinformation are needed to determine clothing requirements and to extend the fuel and lightingrequirements across climate zones. I used household expenditure surveys to set therequirements. In the case of fuel, these are corroborated and extended with engineeringcalculations.

For the expenditure survey approach, I use Prokopovich’s (1909) survey of StPetersburg workers in 1907/8 and Shirras’s (1923) survey of workers in Bombay cotton millsin 1921/2. These were chosen to represent opposite ends of the climate spectrum and becausethe surveys are very early and were taken when the workers were indubitably poor. Bothsurveys show average spending on clothing, footwear, bedding, fuel, and lighting. TheBombay survey breaks all categories down by income bands, and the St Petersburg surveydoes the same for clothing, footwear, and bedding. In Bombay the range 30-40 rupees/monthwas the lowest income band with a large number of workers as was 300-400 rubles/year in StPetersburg. I assume these low income workers were at similar levels of deprivation, so thatdifferences in their expenditures represent responses to climate and not to real income orprice differences.17 For fuel and lighting, the averages for all workers provide a less nuancedbasis of comparison.

The surveys reveal much more substantial purchases of clothing, footwear, andbedding in Russia than in India. Both surveys tell us expenditures in money–rupees orrubles. To compare quantities in the two countries, these must be divided by prices. Forclothing and related items, the prices of coarse cotton cloth were used as the deflator. In thatway we compare expenditures in ‘metres of cloth equivalents’. Table 9 shows the results. InSt Petersburg, the low wage workers consumed almost three times as many metre-equivalentsof clothing, footwear, and bedding as their counterparts in Bombay. Clothing consumptionwas almost 60% greater, bedding was eight times more–the nights are much colder in Russiathan in India–while footwear was, not surprisingly, 27 times greater. Spending on theseitems increased more with income in Russia than in India. The average family member in StPetersburg consumed almost four times as many metre-equivalents as the average Bombayfamily member. Much of the extra income went on clothing for which the Russianconsumption was three times the India. Living in the northern winter required considerablymore clothing.

Similar calculations for lighting and fuel can only be done for the average worker since the Russian survey did not break these expenditures down by income band. Eachmember of the average working class household in Bombay consumed 0.37 million BTUs ofkerosene in lamps (very close to the Millenium Development Goal), while the averagehousehold member in St Petersburg consumed 0.87 million BTUs–over twice as much. Thislooks like the cost of long winter nights.

The disparity was much greater for fuel. In Bombay, fuel consumption averaged 3.15

17Without the adjustments for climate, the real annual earnings of the average cottonmill operative in Russia in 1907/8 look to have been about 50% higher than that of theircounterparts in Bombay in 1921/2. With the climate adjustments, their average real earningswere virtually identical (Allen and Khaustova 2017).

17

million BTUs per person. Among the low income workers, fuel consumption was only 2.52million BTUs, marginally greater than the Millennium Development Goal. In Russia,however, average consumption in working class families was 24.62 million BTUs peryear–close to 10 times more than in Bombay. One limitation of this calculation is that therewas no breakdown of fuel spending by income class in Russia. Judging by clothing,footwear, and bedding, where average spending was double that of the low wage workers, thelow wage workers in St Petersburg might have been consuming on the order of 12 millionBTUs per person.

We can test this conjecture by approaching the problem from a different perspective. Heating engineers have developed a methodology to calculate the energy required to keep abuilding at a desired temperature.18 Critical parameters are the dimensions of the space to beheated, the temperature to be maintained, the pattern of the exterior temperature over theyear, and the insulating efficiency of the construction. No matter how many rooms therewere in a dwelling, it was normal to heat only one, and we proceed accordingly. On theassumption of 3 square metres per person, a family of four lived in a room of 12 squaremetres. The room is assumed to have been 3 x 4 metres with a ceiling height of 2.4 metres. The R value of the floor, walls, and ceiling depends on the construction materials used, theirthickness, and layering. An R value of 2 is assumed.19 We assume the room is heated to aninternal temperature of 15 degrees centigrade. The external temperature is measured by the‘heating-degree days,’ that is, the sum over the year of the difference between the desiredinternal temperature and the external temperature. We obtained this from the heatingindustry website http://www.degreedays.net/. This website gives heating degree-dayscalculated at half hour intervals over five years for most airports and weather stations in theworld. The values chosen for the parameters could be debated, but alternatives give similarresults. Under the assumptions made, the fuel required per person per year works out to havebeen 12 million BTUs in St Petersburg and zero in Bombay. For the latter, the appropriatefuel allowance is the 1.6 million BTUs required for cooking in the Energy Poverty Line ofthe Millennium Development Goals.

Heating degree days were ascertained for major cities in the twenty countries studiedhere, and heating requirements were then calculated. Requirements were set at thesecalculated values so long as they exceeded the Millenium Development energy poverty line;otherwise, the poverty line value of 1.6 million BTUs was adopted. Requirements forlighting, clothing, footwear, and bedding were scaled between those for Bombay and StPetersburg in proportion to the heating degree days of the city relative to the differencebetween Bombay and St Petersburg.

The cost of the requirements for fuel, lighting, clothing, footwear, and bedding can becalculated from the prices of cloth and fuels in the ICP. The cheapest source for each

18http://hyperphysics.phy-astr.gsu.edu/hbase/thermo/heatloss.html summarizes thebasic theory and equations. I am indebted to Michail Moatsos, who has used thismethodology in his own work, for bring it to my attention. See Moatsos (2015).

19For R values of common building materials, see, for instance,http://www.coloradoenergy.org/procorner/stuff/r-values.htm Inspectapedia (2008) assessedan old log cabin with an uninsulated roof, upper exterior walls made of 3/4 inch lumber, anddrafty windows as having an R-value of 2 overall. http://inspectapedia.com/structure/Log_Home_Insulation.php

18

requirement in each country was used in these calculations.

Basic needs poverty line and the extent of world poverty

The basic needs poverty line (BNPL) is the sum of the cost of the linear programmingdiet, the non-food costs, and rent. How do the BNPLs compare to the World Bank povertyline? Does the BNPL change our view of global poverty?

To allow comparison with the WBPL of $1.90 per day, Table 11 shows the values ofthe BNPLs when they are converted to US dollars in the usual manner. The cost of theBNPL line is greater, the higher is the quality of the diet. The BNPL lines for the OECDcountries exceed those of the non-OECD countries for any diet in view of the much highercost of housing in developed countries20 and the relative coldness of the climate. The CPFline at $1.84 comes closest to the WBPL for developing countries. The CPF model could beregarded as micro-foundations for the World Bank’s line. However, since people eating aCPF diet suffer many nutritional deficiencies, it is not a good poverty line. Instead, the lineimplied by the basic diet is preferable. Henceforth, I will confine discussion of the BNPL tothe version using the basic diet.

Table 12 shows a breakdown of poverty line spending by broad category. In thedeveloping countries, about two-thirds of spending is on food, one quarter on non-foods, and5-10% on rent. This rent share is consistent with the experience of pre-industrial Europe(Allen 2001). These shares shift dramatically with income. The food share drops to onequarter in the USA, UK, and France, the non-food share remains at one quarter, and the rentshare explodes to half or more of income. The very poor in the United States pay this muchor more in rent (Desmond 2016).

In Table 13, the expenditures on broad categories of the basic poverty line areconverted to US dollars using both the market exchange rate and the PPP exchange rate forindividual consumption expenditure by households normally used by the World Bank. Theconversion at the market exchange rate shows that food is most expensive in rich countries. The cost of non-food goods is also much greater among the rich than elsewhere due to thecolder climate in the USA and northern Europe, and high rents in rich countries mean thatspending on housing is an order of magnitude greater than in the developing world. Whencurrencies are converted at PPP, costs rise in the poor countries relative to the rich, and foodcosts in dollars end up being greater in the poor countries than in the rich. The burden ofcold weather and the high cost of housing in rich countries remain, although diminished inrelative magnitude.

20The costs per square metre of housing for six of the developing come from ICP2011,as noted, and appear to be ‘nationally representative.’ The costs for the remaining countriesare the costs of poor quality housing in low income districts in large cities. Sensitivityanalysis of the impact of this selection procedure was done to see if it affected the BNPL. Halving the cost of such housing in the developing countries, for instance, had only anegligible impact on the BNPL since even expensive urban housing in those countries is verycheap. Since the prices of everything else are ‘nationally representative,’ the BNPLs for thedeveloping countries also are ‘nationally representative.’ This, however, is not true in the richcountries where the cost of housing amounted to just over half of the cost of the povertybudget. Halving the cost of housing there had a large impact on the BNPL. For the richcountries, the BNPLs must be interpreted as applying to large cities.

19

The basic needs poverty line cost on average about 40% more than the WBPL indeveloping countries (Table 11). The dispersion about this average, moreover, is substantial. The basic diet BNPL in several African countries cost less than $1.90 per day. On the otherhand, the basic diet BNPL cost almost twice as much in some south Asian countries and inthe rich OECD countries.

The import of these discrepancies is that world poverty is greater than implied by theWorld Bank Poverty Line and its geographical distribution is different. Table 14 shows headcount poverty rates for these countries as implied by the $1.90 line and by the basic dietBNPL. The headcounts are computed with the World Bank’s PovcalNet online calculatorand, in the cases of the USA, UK, and France, directly from the national household surveysmade available by the Luxembourg Income Study.21 The BNPL indicates that there were50% more poor people than estimated by the World Bank in the 17 countries in the table forwhich we have estimates. The picture of sub-Saharan Africa does not change much. On theother hand, there are many more poor people in Asia. In India, the increase is 20%, in China74%, and in Indonesia, Vietnam, and Sri Lanka, the number of poor increases by about threefold. The numbers involved are large, and the head count ratio in Indonesia, for instance,rises from 14% to 46%. Economic growth in Asia has been significant but perhaps not asimpressive as it seemed. While the previously noted decline in the consumption of sorghumand millet in India and China suggests the number of poor people is falling, it is stillsubstantial.

Table 14 also reports numbers in absolute poverty in OECD countries. This is a newfocus of research and one of considerable importance in view of current interest in inequalityin these countries. Edin and Schaefer (2015) estimated that 2.8 million children lived on lessthan $2 per day in the USA. Table 11 shows that people cannot survive on $2 a day since itdoes not cover the minimum food, clothing, and shelter required in big cities in a cold, richcountry. Using the basic needs value of $3.72 per day for the USA implies that 1.5% of theAmerican population–4.6 million people–lives in absolute poverty. This is marginally morethan in the UK (1.25%) and significantly more than France (.63%). If France isrepresentative of western Europe, then extreme poverty looks to be a peculiarly Anglo-American problem.

The role of rent in the BNPLs of rich countries is particularly prominent since theirrents are so high. Rents in rural locations in the USA, however, can be very low. In thatcontext, the WBPL offers an insight, for it sets rent close to zero. Even with that reduction,however, there were still 3.6 million people in extreme poverty in the USA in 2011. Countryliving does not eliminate absolute poverty.

Poverty Purchasing Power Parity

One of the contentious issues that has arisen with the World Bank Poverty line is theexchange rate to use in converting the dollar value of the line into local currencies. Thestandard World Bank procedure is to use the purchasing power parity (PPP) exchange rate for

21www.lisdatacenter.org The samples for 2010 were used, and the 2011 poverty lineswere converted to 2010 values with the rates of consumer price inflation in each country. This source was used for USA, UK, and France since the World Bank counsels against usingPovcalNet for these countries. PovcalNet gives similar estimates of the number of poor tothose computed with lisdatacenter.org, but the latter permits finer measurement.

20

individual consumption expenditure by households. The spending pattern of the poor differsfrom the average pattern and so does the spending pattern of developing countries vis-à-visOECD countries. The question is whether a poverty purchasing power parity (PPPP)exchange rate would give a different conversion.

This issue can be explored with the BNPL’s calculated in this paper. In the dualformulation of the standard model of consumer choice, the solution to the problem ofminimizing the cost of reaching a specified utility level is the expenditure function, whichexpresses that cost as a function of the prices of the consumer goods and the specified utilitylevel. If the expenditure function is evaluated at two sets of prices, the ratio of theexpenditures in the two cases is the ‘true cost of living index’: It indicates the relative changein spending needed to compensate the consumer for the differences in prices by keeping himor her at the same level of utility. The solution to the linear programming problem does notgive an explicit expenditure function, but it does indicate the cost of meeting the specifiedrequirements at the given prices. The ratio of the costs of meeting the specified requirementsat two sets of prices is the linear programming analogue to the true cost of living index. Consequently, when the nutritional requirements are set at poverty levels, that ratio is the truepoverty purchasing power parity (PPPP) exchange rate.

It should be noted that the ratio of the costs in the ‘true linear programming cost ofliving index’ is not an index number of the orthodox sort. The numerator and thedenominator need not have any foods in common, for instance. Uniform requirements ensurecomparability as the consumption pattern shifts in response to price differences.

Table 15 shows how the linear programming PPPPs for the various levels ofnutritional requirements compare to the household expenditure PPPs. Among the rich OECDcountries, the linear programming purchasing power parity exchange rates average close tothe exchange rate for household consumption normally used to converted poverty linesbetween countries. Among the middle income OECD countries and the developingcountries, however, the situation is very different. Their PPPP exchange rates are about halfto three quarters of the PPP exchange rates for household consumption. Using the householdconsumption PPP to convert poverty lines gives seriously misleading results.

Conclusion

In this paper we have proposed that poverty lines can be defined by specifying a basicneeds basket. Linear programming is used to specify the food component of basic needs. While the diet problem was the first problem ever formulated as a linear programmingproblem and remains a classic for teaching purposes, the common view amongst economistsis that it does not describe anyone’s behaviour. While that belief is certainly appropriate forrich people, we have argued that it is not correct for the ‘absolute poor’. When people are onthe margin of survival, their needs take precedence over their desires, and their behaviour isgoverned by linear programming. This statement is not unambiguous, however, for a rangeof nutritional requirements can be imposed on the diet problem. We have argued that the‘basic’ requirements–those which supplied adequate amounts of calories, protein, fat, and thevitamins and minerals needed to prevent anaemia, beri-beri, pellagra, and scurvy–imply dietsthat describe the main features of the diets of the poor. Those diets are based aroundcommon grains, legumes, a little milk or fish, oil, and vegetables. Linear programmingcannot, however, describe all of the details of the diets.

We have also explicitly budgeted the non-food component of ‘basic needs’ andexpressed many of them as functions of climate. This is important if the poverty line is to be

21

applicable everywhere on the globe since the existing World Bank procedure gives a linewhich is only appropriate for hot climates. Accomodation rents vary enormously betweenrich and poor countries, and the basic needs line includes that differential, while the WorldBank line does not.

Although the poverty line proposed here is intended to be austere, it does nonethelessprovide some latitude for habit and taste since the nutritional requirements are thoserecommended by the medical profession for ‘good health.’ This leaves scope for people totrade off some health for customary consumption, sugar, or alcohol, as they prefer. Basicneeds are not the same as local taste. However, the trade-offs are limited. That is the natureof poverty. We argued in the introduction that an international poverty line should satisfy fiveconditions. The basic needs poverty line meets all of them. First, the line should have a clearmeaning related to survival. The BNPL meets this condition since it is defined in terms ofthe food, fuel, clothing, and shelter requirements to ensure social reproduction and defenceagainst the main deficiency diseases and survival in cold as well as hot countries. Second,the line should represent a constant standard across time and space. This requirement is metby imposing the same or equivalent requirements in all cases. Third, the poverty line shouldrespond to local prices and climate. Indeed, local prices determine the solution to the linearprogramming along with the nutritional requirements, and non-food consumption varies withclimate. Fourth, the poverty line should avoid intractable index number problems. There areno index number problems with the linear programming approach since the solutions to thediet problem are in local prices and the non-food requirements are also costed with localprices. Comparability across countries and over time is guaranteed by using the samerequirements everywhere–not by PPP. Fifth, the poverty line should require only readilyavailable information. An ICP data set including relevant accommodation costs would be dothe job, although it is not detailed enough for regional breakdowns in large countries.

The basic needs poverty line provides a direct connection between the value of theline and its meaning in terms of human health and social reproduction. Using the BNPLprovides a more transparent approach to poverty measurement than existing World Bankprocedures. The BNPL indicates that there is considerably more poverty in the countriesanalysed here–and they include much of the population of the developing world–than theWorld Bank has counted. The BNPL also indicates there are millions in ‘absolute poverty’ inrich countries–-especially the USA and UK.

22

Table 1

1700 Calorie model diets(kilograms per person per year)

23

Table 2

Total weight of linear programming diets(kilograms per person per year)

24

Table 3

Number of items in linear programming diets

25

Table 4

CPF diets(Kilograms per person per year)

26

Table 5

CPF diet: vitamins and minerals relative to RDA

27

Table 6Basic diets

(Kilograms per person per year)

28

Table 7Full course diets

(Kilograms per person per year)

29

Table 8LP predictions compared to 1961 average consumption

(Kilograms per person per year)

total grain/ fats/ animal veg, nuts weight bread oils & fish & fruit other

eleven developing countriespredicted 293 170 5 28 90 01961 average 334 138 4 41 132 19

low income OECDpredicted 312 192 3 70 47 01961 average 593 183 9 147 222 32

rich OECDpredicted 314 203 12 48 51 01961 average 906 109 17 358 259 164

30

Table 9

Non-food consumption per head among Workers in Bombay and St Petersburg

Bombay St Petersburg Low wage Average Low wage Average

clothing, footwear, bedding (in cotton cloth equivalents,metres)

clothing 17.00 23.13 26.86 62.50footwear .59 1.19 16.19 30.63bedding 1.28 3.38 10.08 21.37 total 18.88 27.69 53.13 114.50

fuel (mBTU) 2.52 3.15 24,62

light (mBTU) .27 .37 .87

Source: Prokopovich (1909)and Shirras (1923).

31

Table 10

Housing rents

32

Table 11

Linear Program Poverty Lines converted to US dollars per dayat PPP

33

Table 12Expenditure Breakdown of Basic Needs Poverty Line

34

Table 13

Expenditures by Category in US Dollarsat market and PPP exchange rates

35

Table 14

Head Count Poverty Rates

Note: PovcalNet does not generate poverty estimates forAlgeria, Egypt, or Myanmar. The figures for the UK, USA, andFrance were computed from the 2010 household surveys in theLuxembourg Income Survey Cross-National Data Centrehttp://www.lisdatacenter.org/

36

Table 15

Linear Programming PPPP relative to Individual Consumption PPP

37

References

Ahluwalia, M.S., Carter, N.G., and Chenery, H.B. (1979). “Growth and Poverty inDeveloping Countries,” Journal of Development Economics, Vol. 6 , pp. 299-341.

Aichholz, Karsten (2016). Renting in Bangkok: Locations, Prices and Contracts,www.thailandstarterkit.com/accomodation/bangkok-apartment.

Allen, Robert C. (2001). "The Great Divergence in European Wages and Prices from theMiddle Ages to the First World War," Explorations in Economic History, Vol. 38, October,pp 411-447

Allen, Robert C. (2011). Global Economic History: A Very Short Introduction, Oxford,Oxford University Press.

Allen, Robert C. (2013). “Poverty Lines in History, Theory, and Current InternationalPractice,” Oxford University, Department of Economics, Working Paper 685.

Allen, Robert C., and Khaustova, Ekaterina (2017). “Russian Real Wages Before and After1917: In Global Perspective” Oxford University Working Papers in Economic and SocialHistory, No. 158.

Atkin, David (2013). “Trade, Tastes, and Nutrition in India,” American Economic Review,Vol. 103, pp. 1629-63.

Atkin, David (2016). “The Calorie Costs of Culture: Evidence from Indian Migration,”American Economic Review, Vol. 106, pp. 1144-81.

Atkinson, Antony (2017). Monitoring Global Poverty: Report of the Commission on GlobalPoverty, Washington, The World Bank.

Barnes, Douglas F., Khandker, Shahidur R., and Samad, Hussain A. (2011). “EnergyPoverty in Rural Bangladesh,” Energy Policy, Vol. 39, pp. 894-904.

Blackorby, Charles and Donaldson, David. (1987). “Welfare ratios and distributionallysensitive cost-benefit analysis,” Journal of Public Economics, Vol. 34, pp. 265–90.

Bombay, Labour Office (1928). Report on an Enquiry into Working Class Family Budgets inAhmedabad, Bombay, Government Central Press.

Carlson, A., Lino, M., Juan, W-Y., Hanson, K., & Basiotis, P.P. (2007). Thrifty FoodPlan, 2006. (CNPP-19). U.S. Department of Agriculture, Center for Nutrition Policyand Promotion.

Chen, S., and Ravallion, M. (2001). “How did the World’s Poorest Fare in the 1990s?”Review of Income and Wealth, series 47, No. 3, pp. 283-300.

Chen, S., and Ravallion, M. (2010). “The Developing World is Poorer than We Thought, But

38

No Less Successful in the Fight Against Poverty,” The Quarterly Journal of Economics, Vol.125 (4), pp. 1577-1625.

Deaton, Angus (2010). “Price Indexes, Inequality, and the Measurement of World Poverty,”American Economic Review, Vol. 100, pp. 5-34.

Deaton, Angus, and Drèze, Jean (2009). “Food and Nutrition in India: Facts andInterpretations,” Economic and Political Weekly, Vol. 44, pp. 42-65.

Deaton, Angus, and Heston, Alan (2010). “Understanding PPPs and PPP-based NationalAccounts,” American Economic Journal: Macroeconomics, Vol. 2, pp. 1-35.

Deaton, Angus, and Kozel, Valerie (2005). The Great Indian Poverty Debate, Delhi,Macmillan India Ltd.

Desmond, Matthew (2016). Evicted, New York, Crown Publishers.

Edin, Kathryn J., and Shaefer, H. Luke (2015). $2 a Day: Living on Almost Nothing inAmerica, Boston, Mariner Books.

Ferreira, Francisco H.G. et al. (2016). “A Global Count of the Extreme Poor in 2012: DataIssues, Methodology and Initial Results,” Journal of Economic Inequality, vol. 14, pp. 141-72.

Food and Agricultural Organization of the United Nations (2001). Human EnergyRequirements: Report of a Joint FAO/WHO/UNU Expert Consultation

Food and Agricultural Organization of the United Nations (2008a). The State of FoodInsecurity in the World

Food and Agricultural Organization of the United Nations (2008b). FAO Methodology forthe Measurement of Food Deprivation: Updating the minimum dietary energy requirements.

Food and Agricultural Organization of the United Nations (2008c). Fats and Fatty Acids inHuman Nutrition: Report of an Expert Consultation.

Jamal Vali (2001). “Chasing the Elusive Rural-Urban Gap in Tanzania,” Journal ofContemporary African Studies, Vol. 19, No. 1, pp. 25-38.

Godanho, Pedro (2014). Uneven Growth: Tactical Urbanism for Expanding Megacities,New York, The Museum of Modern Art.

Goedemé, Tim, Bérénice Storms, Tess Penne, Karel Van den Bosch, eds, (2015). PilotProject for the Development of a Common Methodology on Reference Budgets inEurope–Final Report, Brussels, European Commission.

International Energy Agency, Energy Prices and Taxes: Quarterly Statistics, 2012, fourthquarter.

39

Jamal Vali (2001). “Chasing the Elusive Rural-Urban Gap in Tanzania,” Journal ofContemporary African Studies, Vol. 19, No. 1, pp. 25-38.

Jinqing, Cao (2005). China Along the Yellow River: Reflections on Rural Society, trans yNicky Harman and Huang Ruhua, London, RoutledgeCurzon.

Jolliffe, D., and Prydz, E. (2016). “Estimating International Poverty Lines from ComparableNational Thresholds,” Journal of Economic Inequality, Vol. 14, pp. 185-98.

Kakwani, N., and Son, H. (2016). “Global Poverty Measurments Based on 2011 PurchasingPower Parity: Where Should the New Poverty Line be Drawn?” Journal of EconomicInequality, Vol. 14, pp. 173-84.

Kander, Astrid (2002). Economic Growth, Energy Consumption, and CO2 Emissions inSweden: 1800-2000, Stockholm, Almqvist & Wicksell International.

Katsoulakos, Nicholas M., and Kaliampakos, Dimitris C. (2014). “What is the impact ofaltitude on energy demand? A Step towards developing energy policy for mountainous areas,Energy Policy, Vol. 71, pp. 130-8.

Lindgren, Mattias (2015). “The Elusive Quest for the Subsistence Line: How much does thecost of survival vary between populations?” Munich Personal RePec Archive, Paper Number73891.

Lockyer, Charles (1711). An Account of the Trade in India, London.

Lynch, F. T. (2012). The Book of Yields – Accuracy in Food Costing and Purchasing., 8th

edition, Hoboken, N.J, John Wiley & Sons.

Luxembourg Income Study (LIS) Database, http://www.lisdatacenter.org (France, USA,United Kingdom; {Acessed 27 June 2017}). Luxembourg: LIS.

Moatsos, Michail (2015), “Global Absolute Poverty: Behind the Veil of Dollars,” Universityof Utrecht.

Modi, V., S. McDade, D. Lallement, and J. Saghir (2006). Energy and the MillenniumDevelopment Goals, New York, Energy Sector Management Assistance Programme, UnitedNations Development Programme, UN Millennium Project, and World Bank.

Musgrove, Philip (1984). “Household Income Distribution in the Dominican Republic: AFirst Look at New Survey Data,” in The Collection and analysis of Economic and ConsumerBehavior Data: In Memory of Robert Ferber, Bureau of Economic and Business Research &Survey Research Laboratory, University of Illinois, pp. -121.

Musgrove, Philip (1986). “Distribución del Ingreso Familiar en la República Dominicana,1976-1977: La Encuesta Nacional de Ingresos y Gastos Familiares,” El Trimestre económico,Vol. 53, pp. 341-92.

40

Oshodi, Lookman (2011). “Housing and Urban Development in Monrovia,Liberia–Observer’s Perspective.”

Pokharel, Shaligram (2004). “Energy Economics of Cooking in Households in Nepal,”Energy, Vol. 29, pp. 547-59.

Prinzo, Zita Weise (2000). Pellagra and its Prevention and Control in Major Emergencies,World Health Organization, WHO/NHD/00.10.

Prokopovich, S.N. (1909). Biudzhety petersburgskikh rabochik, St. Petersburg.

Rao, B. S. Narasinga (2009). Nutrient Requirements and Recommended Dietary Allowancesfor Indians: A Report of the Expert Group of the Indian Council of Medical Research,Hyderabad, National Institute of Nutrition.

Ravallion, M. (2010) “How not to count the poor: A Reply to Reddy and Pogge.” in Debatesin the Measurement of Poverty, ed. By Joseph Stiglitz, Sudhir Anand and Paul Segal, OxfordUniversity Press, pp. 86-101

Ravallion, M. (2016). “Are the World’s Poorest Being Left Behind,” Journal of EconomicGrowth, Vol. 21, pp. 139-64.

Ravallion, M., Datt, G., van de Walle, D. (1991). “Quantifying Absolute Poverty in theDeveloping World,” Review of Income and Wealth, series 37, No. 4, pp. 345-61.

Ravallion, M., Datt, G., van de Walle, D., Chan E. (1991). “Quantifying and Severity ofAbsolute Poverty in the Developing World,” PRE Working Paper 587, World Bank.

Ravallion, M. and Chen, S. (1997). “What can New Survey Data Tells Us about RecentChanges in Distribution and Poverty?” The World Bank Review, Vol. 11, No. 2, pp. 357-82.

Ravallion, M., Chen, S., and Sangraula, P. (2009). “Dollar A Day Revisited,” World BankEconomic Review, Vol. 23, No. 2, pp. 163-84.

Reddy, S.G., and Pogge, T. (2010). “How not to count the poor,” in Debates in theMeasurement of Poverty, ed. By Joseph Stiglitz, Sudhir Anand and Paul Segal, OxfordUniversity Press, pp. 42-85.

Roberts-Hughes, Rebecca (2011). The Case for Space, Royal Institute of British Architechts.

Sabry, Sarah (2010). “How Poverty is Underestimated in Greater Cairo, Egypt,” Environmentand Urbanization, Vol. 22, pp. 523-41.

Sen, Amartya (1987). “The Standard of Living,” in Sen, Muellbauer, Kanbur, Hart, andWilliams, The Standard of Living: The Tanner Lectures on Human Values, Cambridge:Cambridge University Press.

Sen, Armartya (1999). Development as freedom. New York, Oxford University Press.

41

Sen, Amartya (1992). Inequality Re-examined, Oxford: Clarendon Press.

Shirras, G. Findlay (1923). Report on an Enquiry into Working Class Budgets in Bombay,Bombay, Government Central Press.

Siddiqui, Danish (2015). “From Shanty to Swanky: the wildly different price of a property inMumbai–in pictures,” The Guardian,https://www.theguardian.com/cities/gallery/2015/apr/23/price-property-mumbai-in-pictutes.

Sillers, D. (2015). “Is $1.82 the new $1.25? Choosing the Next International ExtremePoverty Line,” USAID Economics Brief, Washington, U.S. Agency for InternationalDevelopment.

Smith, Adam (2007). An Inquiry into the Nature and Causes of the Wealth of Nations,Amsterdam, Metalibri, http://metalibri.incubadora.fapesp.br, accessed 23 January 2016.

Smith, Victor E. (1959). “Linear Programming Models for the Determination of PalatableHuman Diets,” Journal of Farm Economics, Vol. 41, pp. 272-83.

Stigler, George J. (1945). “The Cost of Subsistence,” Journal of Farm Economics, Vol. 27,pp. 303-14.

Sun, Linyun (2011). In Search of an Affordable Housing System for Shanghai, China, Thesisfor Master of Science in Real Estate Development, MIT.

Suryanarayana, M.H. (2009). “Nutritional Norms for Poverty: Issues and Implications,”Concept paper prepared for the Expert Group to Review the Methodology for Estimation ofPoverty.

Unger, Richard W., and Thistle, John (2013). Energy Consumption in Canada in the 19th and20th Centuries: A Statistical Outline, Naples, Consiglio Nazionale delle Richerche.

UN-Habitat (2003a). Rental Housing: An Essential Option for the Urban Poor inDeveloping Countries, United Nations Human Settlements Program, Nairori, 2003.

UN-Habitat (2003b). The Challenge of Slums: Global Report on Human Settlements, 2003,UN Human Settlements Program, London, Earthscan Publications, Ltd.

UN-Habitat (2010). The Challenge of Slums: Global Report on Human Settlements, 2003,Chapter 1: Development Context and the Millennium Agenda, revised and updated version,http://unhabitat.org/wp-content/uploads/2003/07/GRHS_2003_Chapter_01_Revised_2010.pdf

United States Department of Agriculture. (2010). Food Security Assessment, 2010-2020,

42

Appendix

Notes on the linear programming

The following foods were represented in the objective function (equation 1) in the linearprograms:

wheat flour, wheat bread, rice, maize flour, oatmeal, beans, eggs, cheese, chicken, milk,meat, fish, butter, margarine, vegetable oil, white sugar, potatoes, tomatoes, cucumbers,cabbage, sweet potatoes, carrots, onions, cauliflower, spinach, ginger, roasted peanuts,avocado, mushrooms, cassava, tofu, alcohol

In addition, the programs for Africa also included millet, millet flour, red sorghum grains,yellow maize grains, and white maize grains

When prices for items in this list were not available in a country, they were excluded fromthe choice options in the linear program.

Generally, the nutritional composition of foods are coded per ‘edible portion.’ In therequirement inequalities in the linear programs (equation 2), the nutritional compositionsreported in the databases like the USDA National Nutrient Database were multiplied by thepercentage of the food that is edible (ie allowing for bones, etc), so that the solution of thelinear program was the weight of food purchased at the prices specified in ICP2011.

The ICP contains many foods, most of which are clearly too expensive to be chosen by thelinear program. For instance, thirteen different kinds of fresh fish or seafood are listed in thecore price list of ICP2011, although not all prices are available in all countries. The fish thatwas cheapest in terms of its nutritional content (allowing for losses in bones, fins, etc) wasused in the linear program. This was usually mackerel or tilapia or carp. Similar selectionswere made for meat, bread, and alcohol in terms of alcohol content. The African regionalprice list contains a great variety of local alcoholic beverages, which were also investigatedto find the cheapest source of alcohol.

Sometimes ICP2011 reported prices for goods that could not have been widely available. Thus, there is a price for mackerel in Zimbabwe. At that price, mackerel is included in theleast cost diet. It is so implausible, however, that mackerel is widely available in Zimbabwethat it was excluded from the choice options for that country. Similar exclusions were madein some other cases when it was deemed that the food in question was not widely available.

43

Online data sources

While the ICP is a tremendous achievement, it is not complete, so it was necessary toadd missing variables derived from extraneous sources. Some important additions included:! the price of wheat flour. Wheat flour is of great importance in poverty measurement,

but it was curiously missing from the data for the United States. This was particularlyimportant since conversion to US dollars is a key part of the World Bank’s exercise. Why the price is missing is altogether puzzling since it is available on the US Bureauof Labor Statistics website.22 That price was used in the calculations reported here. Wheat flour prices were also taken from the national online retail price databases forsome other countries.

! the price of cabbage. This turns out to be important since it is the cheapest source ofvitamin C in many places. It was also missing from the USA data in the ICP2011 andthe price was taken from the BLS database. In the case of some countries like theUK, the price of cabbage in 2011, which was missing, was estimated from relativeprices in 2015 taken from supermarket databases. The same relatives were assumedto obtain in 2011.

! some missing USA and UK prices (like toilette soap, maize flour, and oatmeal in theUSA) were taken from data collected in 2011 from supermarket web sites for earlierinvestigations.

! the price of electricity. This was necessary for the non-food component of theproblem. ICP2011 lacks electricity prices for many south and east Asian countries. These were taken from the online tariffs of their electricity suppliers23. Informationfor some countries in 2011 was provided by Mr. Beni Suryadi of the ASEAN Centrefor Energy, and his help is grateful acknowledge.

! The price of kerosene was missing for many OECD countries. Prices of kerosene or‘light fuel oil for households’ were taken from IEA (2012).

! the rental price per square metre of housing. ICP 2011 inquired about rents, but dataare available for only six countries in the sample (Algeria, Niger, Bangladesh,Indonesia, Sir Lanka, and Vietnam.) For these the price per square metre of a ‘typicaltraditional dwelling,’ generally supplied with electricity and indoor water, but noprivate toilet, private kitchen, or air conditioning, was used. For other countries, rentswere obtained for low standard accommodation in poor districts in large or capitalcities was obtained.

Food Composition web sites

United States Department of Agriculture (USDA). National Nutrient Database for StandardReference Release 28. Retrieved on 26 January 2016 from http://ndb.nal.usda.gov/ndb/search

22http://www.bls.gov/regions/mid-atlantic/data/AverageRetailFoodAndEnergyPrices_USandMidwest_Table.htm

23See appendix of online sources for websites consulted.

44

Health Canada. Canadian Nutrient File. Retrieved on 26 January 2016 fromhttp://webprod3.hc-sc.gc.ca/cnf-fce/start-debuter.do?lang=eng

Stadlmayer, B. et al. (Eds.). Food and Agriculture Organization of the United Nations (2010).Composition of Selected Foods from West Africa. Retrieved on 26 January 2016 fromhttp://www.langual.org/langual_linkcategory.asp?CategoryID=4&Category=Food+Composition

Stadlmayer, B. et al. (Eds.). Food and Agriculture Organization of the United Nations (2012).West African Food Composition Table. Retrieved on 26 January 2016 fromhttp://www.langual.org/langual_linkcategory.asp?CategoryID=4&Category=Food+Composition

Latvia, Food and Veterinary Service. Latvian Food Composition Database. Retrieved on 26January 2016 from http://www.partikasdb.lv/

French Agency for Food, Environmental and Occupational Health and Safety. French FoodComposition Table 2012. Retrieved on 26 January 2016 fromhttp://www.afssa.fr/TableCIQUAL/

ASEAN Food Composition Database (2014). Retrieved on 26 January 2016 fromhttps://www.google.ae/url?sa=t&rct=j&q=&esrc=s&source=web&cd=2&cad=rja&uact=8&ved=0ahUKEwjTi72Vv8fKAhWsn4MKHUH5ASIQFggiMAE&url=http%3A%2F%2Fwww.inmu.mahidol.ac.th%2Faseanfoods%2Fdoc%2FOnlineASEAN_FCD_V1_2014.pdf&usg=AFQjCNFNSHKYTMVzGG0LPEhv-E_mbxfl6g&sig2=U436wr8h7O1dnK37dOFVzQ

Nutritional RequirementsU.S. Food and Drug Administration. Guidance for Industry: A Food Labeling Guide (14.Appendix F: Calculate the Percentage Daily Value for the Appropriate Nutrients). Retrievedon 26 January 2016http://www.fda.gov/Food/GuidanceRegulation/GuidanceDocumentsRegulatoryInformation/LabelingNutrition/ucm064928.htm

National Institute of Nutrition. Indian Council of Medical Research (2009). NutrientRequirements and Recommended Dietary Allowances for Indians. Retrieved on 26 January2016 fromhttps://www.google.ae/url?sa=t&rct=j&q=&esrc=s&source=web&cd=1&cad=rja&uact=8&ved=0ahUKEwjDkYPRmcfKAhVLuIMKHYW0B08QFggbMAA&url=http%3A%2F%2Ficmr.nic.in%2Ffinal%2Frda-2010.pdf&usg=AFQjCNHrGPR00NBuF-gRaiqurV1ESqTriQ&sig2=PsbdGjHhbh6zPY7dtoHk9g

United Nations. Department of Economic and Social Affairs. Population Division. Retrievedon 26 January 2016 http://esa.un.org/unpd/wpp/Excel-Data/population.htm

Vegetable (and other foods) consumption

United States Department of Agriculture. Daily Vegetable Chart. Retrieved on 26 January2016 from http://www.choosemyplate.gov/vegetables

45

Sachdeva, S., Sachdev, T. R., & Sachdeva, R. (2013). Increasing Fruit and VegetableConsumption: Challenges and Opportunities. Indian Journal of Community Medicine :Official Publication of Indian Association of Preventive & Social Medicine, Vol. 38, No. 4,pp. 192–197.

Hall, J. N. et al. (2009). Global variability in fruit and vegetable consumption. AmericanJournal of Preventive Medicine, Vol. 36, No. 5, pp. 402-409.

Radhika, G. et al. (2011). Dietary profile of urban adult population in South India in thecontext of chronic disease epidemiology. Public Health Nutrition. Vol. 14, No. 4, pp. 591-8.

Food and Agriculture Organization of the United Nations and World Health Organization(2003). Diet, Nutrition and the Prevention of Chronic Diseases. Report of a Joint WHO/FAOExpert Consultation. Retrieved on 26 January 2016 fromhttp://www.fao.org/docrep/005/ac911e/ac911e05.htm

Chauvin, N. C. et al. (2012). Food Production and Consumption Trends in Sub-SaharanAfrica: Prospects for the Transformation of the Agricultural Sector. United NationsDevelopment Programme, Working Paper 2012-011, Feb. Retrieved on 26 January 2016from http://www.africa.undp.org/content/rba/en/home/library/working-papers/food-production-consumption-trends.html

United States Department of Agriculture. Profiling Food Consumption in America. Retrievedon 26 January 2016 fromhttps://www.google.ae/url?sa=t&rct=j&q=&esrc=s&source=web&cd=2&cad=rja&uact=8&ved=0ahUKEwiM24CivcfKAhWMloMKHSsaAAoQFgggMAE&url=http%3A%2F%2Fwww.usda.gov%2Ffactbook%2Fchapter2.pdf&usg=AFQjCNFBcXq4l5gn-vfY0ooT_Mormu0DqQ&sig2=TS4KqjedUUVrKj_-nCWv3Q

Food FortificationFood Fortification Initiative. Retrieved on 26 January 2016 fromhttp://www.ffinetwork.org/index.html

Future of Flour – A compendium of Flour Improvement (2006). Retrieved on 26 January2016 from http://muehlenchemie.de/english/know-how/future-of-flour.html

Electricity pricesAbeberese, A. B. (2013). Electricity Cost and Firm Performance: Evidence from India,Columbia University Academic Commons, http://hdl.handle.net/10022/AC:P:20773.

Central Electricity Authority, Ministry of Power, Government of India. Tariff & Duty ofElectricity Supply in India. Retrieved on 26 January 2016 fromhttp://www.slideshare.net/ashishverma061/tariff-and-duty-of-electricity-supply-in-various-state-of-indiaa-review-by-cea

Nepal Electricity Authority Annual Report. Retrieved on 26 January 2016 fromhttps://www.google.ae/url?sa=t&rct=j&q=&esrc=s&source=web&cd=1&cad=rja&uact=8&ved=0ahUKEwjJlcS8scfKAhUJgYMKHV98DZ0QFggaMAA&url=http%3A%2F%2Fwww.n

46

ea.org.np%2Fimages%2Fsupportive_docs%2FAnnual%2520Report-2011.pdf&usg=AFQjCNFOIM6GXbxG0i5RDo90hVaHANYhbQ&sig2=jy3s5HRsPM6Fu2MN0IUwPA&bvm=bv.112454388,d.cWw

Public Utilities Commission of Sri Lanka. Decision on Electricity Tariffs (Effective from 1st

January 2011). Retrieved on 26 January 2016 from http://www.pucsl.gov.lk/english/wp-content/themes/pucsl/pdfs/decision_on_electricity_tariffs-2011.pdf

Department of Electrical Services, Brunei Darussalam. Retrieved on 26 January 2016 fromhttp://www.des.gov.bn/bdes/index.php?option=com_content&task=view&id=39&Itemid=44

Tariff of EDC for Phnom Penh and Kandal Province. Retrieved on 26 January 2016 fromhttp://www.eac.gov.kh/pdf/report/report%202012%20(en).pdf

Tarif Tenaga Listrik Untuk Keperluan Pelayan Sosial Berlaku Mulai 1 Mei 2014. Retrievedon 26 January 2016 from http://www.pln.co.id/dataweb/Permen-09-2014.pdf

Electricity Tariff. Lao PDR. Retrieved on 26 January 2016 fromhttp://www.edl.com.la/en/page.php?post_id=6

Pricing and Tariff. Malaysia. Retrieved on 26 January 2016 fromhttp://www.sarawakenergy.com.my/index.php/residential

Electricity Tariff. Myanmar. Retrieved on 26 January 2016 fromhttp://www.moep.gov.mm/index.php?option=com_content&view=article&id=66&Itemid=97

Effective Rates. Philippines. Retrieved on 26 January 2016 fromhttp://www.napocor.gov.ph/index.php/2013-09-13-01-23-51/psalm-effective-rates

Historical Electricity Tariff 2010-2014. Singapore. Retrieved on 26 January 2016 fromhttp://www.singaporepower.com.sg/irj/go/km/docs//wpccontent/Sites/SP%20Services/Site%20Content/Tariffs/documents/Historical%20Electricity%20Tariff.xls

Tariff. Thailand. Retrieved on 26 January 2016 fromhttp://www.eria.org/events/Power%20Tariff%20Structure%20in%20Thailand.pdf

Domestic Electricity Retail Price. Vietnam. Retrieved on 26 January 2016 fromhttp://www.moit.gov.vn/Images/FileVanBan/_TT19-2013-BCTeng.pdf

International Institute for Sustainable Development (2014). Indonesia Energy SubsidyReview. Issue 1, Vol. 1, March. Retrieved on 26 January 2016 fromhttps://www.iisd.org/gsi/sites/default/files/ffs_indonesia_review_i1v1.pdf

Nepal Electricity Authority Annual Report (2011). Retrieved on 26 January 2016 fromhttps://www.google.ae/url?sa=t&rct=j&q=&esrc=s&source=web&cd=1&cad=rja&uact=8&ved=0ahUKEwi7-q7VuMfKAhVFg4MKHeBYAuIQFggaMAA&url=http%3A%2F%2Fwww.nea.org.np%2Fimages%2Fsupportive_docs%2FAnnual%2520Report-2011.pdf&v6u=https%3A%2F%2Fs-

47

v6exp1-ds.metric.gstatic.com%2Fgen_204%3Fip%3D91.230.41.194%26ts%3D1453810092914715%26auth%3D4bywyizitxoa6voeaqzaw4hw4t4tfdyt%26rndm%3D0.1636814004741609&v6s=2&v6t=11505&usg=AFQjCNFOIM6GXbxG0i5RDo90hVaHANYhbQ&sig2=BgjWMkYw1fTpCGNkA32-YQ&bvm=bv.112454388,d.cWw

The World Bank. Rethinking Electricity Tariffs and Subsidies in Pakistan. WB ReportNumber 62971-PK. Retrieved on 26 January 2016 fromhttps://www.google.ae/url?sa=t&rct=j&q=&esrc=s&source=web&cd=1&cad=rja&uact=8&ved=0ahUKEwjb2KWRucfKAhXqnYMKHerWCjgQFggaMAA&url=https%3A%2F%2Fopenknowledge.worldbank.org%2Fbitstream%2Fhandle%2F10986%2F19456%2F629710ESW0whit0020110Final00PUBLIC0.pdf%3Fsequence%3D1&usg=AFQjCNG4jc7GbQlGQ08Akn4nyHx2y4oYjA&sig2=iIfCLV23rqz6FtyNlurubQ&bvm=bv.112454388,d.cWw

Energy Regulatory Commission of Thailand (2012). Power Tariff Structure in Thailand.Retrieved on 26 January 2016 fromhttps://www.google.ae/url?sa=t&rct=j&q=&esrc=s&source=web&cd=1&cad=rja&uact=8&ved=0ahUKEwiq7pHiucfKAhUFyYMKHVYfB4IQFggaMAA&url=http%3A%2F%2Fwww.eria.org%2Fevents%2FPower%2520Tariff%2520Structure%2520in%2520Thailand.pdf&usg=AFQjCNGzLUAsVpU3lLEkvGqW9xO2iJB6Rw&sig2=KNEANnX2K2kDF4tcS0q3Dw&bvm=bv.112454388,d.cWw

National food consumption data

Food and Agriculture Organization of the United Nations Statistics Division. Food BalanceSheets. Data retrieved on 26 January 2016 from http://faostat3.fao.org/download/Q/QC/E

PricesMaisto produktu kainos. Retrieved on 26 January 2016 fromhttp://www.produktukainos.lt/?mid=95&archive=1&limit=20&offset=220

Statistics Canada. Food and Other Selected Items, Average Retail Price. Retrieved on 26January 2016 from http://www.statcan.gc.ca/tables-tableaux/sum-som/l01/cst01/econ155a-eng.htm

National Institute of Statistics and Economic Studies. Average Consumer Prices inMetropolitan France. Retrieved on 26 January 2016 from http://www.insee.fr/en/bases-de-donnees/bsweb/theme.asp?id=06

U. S. Bureau of Labor Statistics, Average Retail Food and Energy Prices, U.S. and MidwestRegion, retrieved on 1 February 2016,http://www.bls.gov/regions/mid-atlantic/data/AverageRetailFoodAndEnergyPrices_USandMidwest_Table.htm

Bureau of Labor Statistics. Average Price Data. Retrieved on 26 January 2016 fromhttp://www.bls.gov/data/

Statistics Norway. 250 Retail Prices of Selected Commodities. Retrieved on 26 January 2016

48

from http://www.ssb.no/a/english/aarbok/tab/tab-250.html

"Countries Compared by Cost of living > Prices at markets > 1kg. International Statistics atNationMaster.com", Numbeo.com. Cost of living. Aggregates compiled by NationMaster. Retrievedon 26 January 2016 from http://www.nationmaster.com/country-info/stats/Cost-of-living/Prices-at-markets

Food and Agricultural Organization, Food Price Monitoring and Analysis Tool, retrieved on29 January 2016, http://www.fao.org/giews/pricetool/

India www.indiastat.com

Housing rents

The rental price per square metre of housing was needed to compute the cost of thesubsistence requirement for housing of 3 square metres per person.

For Algeria, Niger, Bangladesh, Indonesia, Sir Lanka, and Vietnam, rents werecalculated from ICP2011 core price list. The core list indicates rental prices for a variety ofunits. I used the price per square metre of a ‘typical traditional dwelling,’ generally suppliedwith electricity and indoor water, but no private toilet, private kitchen, or air conditioning. These are apparently ‘nationally representative.’

Rents for other countries were derived from other sources. The aim was to find rentsfor low quality accommodation in poor districts in large or capital cities. It was not alwayspossible to find 2011 rents, in which case, a rental price index was used to work out a 2011value. The following sources were consulted:Zimbabwe–African Housing Finance Yearbook 2015, p. 213. Rent is for 2014 and wasassumed to apply to 2011 due to overall price stability.Gambia–assumed to be $.50/sq metre/month, which is typical of such a poor country.Liberia–Oshodi (2011, p. 7) Single room without sanitation facilities rents for $10 per monthat low end of distribution. Assumed to be 10 square metres.Egypt–Sabry (2010).India–rent per square foot of a shanty in Dharavi slum, Siddiqui (2015)China–Logan, Fang, and Zhang (2009, p. 921) average cost per square metre at ‘marketrental’ in eight largest cities in China from 2000 census. Adjust to 2011 prices with urbanhousing rental index component of Chinese consumer price index from www.stats.gov.cnThailand–Aichholz (2016, p.7) deflated with index on page 5.Myanmar–assumed to be $.50/sq metre/month, which is typical of such a poor country. Deflated to 2011 with the housing component of India consumer price index.

Turkey–price per square metre from rent of 120 sq metre apartment in Kadikoy district inIstanbul in 2016 from www.globalpropertyguide.com/Europe/Turkey/Rental-Yields Deflated to 2011 with rental price index for Istanbul in numbeo.com.

Mexico–assumed to be $5 per square metre per month.

Lithuania–rent per square metre in Paneriai district of Vilnius, the most impoverished, fromvddv.library.lt

49

USA–rent taken to be $1250 per month for a one bedroom slum apartment of 750 square feet(Godanho 2014). The rent is very close to that give in US Department of Housing, FairMarket Rent database for New York County, NY in fiscal year 2011.

UK–rent for one bedroom flat, lower quartile, in Bexley, the cheapest London district, in2011 from www.gov.uk/government/statistics/private-rental-market-statistics-england-onlyFlat assumed to be 45 square metres following Roberts-Hughes (2011, pp. 23).

France–rent per square metre in 2015 from cheapest category apartment in 18th

arrondissement in Paris from www.lacoteimmo.com/prix-de-l’immo/location/ile-de-france/paris/paris-18eme/750118.htm Deflated from2011 with price index for one bedroomflat outside of city centre for Paris in numbeo.com